The EUMETSAT AVHRR and SLSTR AMV Tropical Speed Biases Pattern Explained
Since EUMETSAT started deriving Atmospheric Motion Vectors (AMVs) from the Advanced Very-High-Resolution Radiometer (AVHRR) on board the Metop satellites, AMVs from EUMETSAT’s Low-Earth Orbit (LEO) platforms are known to have significant speed biases in the tropical band, when compared to the forecast model. LEO AMVs are noticeably too fast at most longitudes in the tropics, while being mostly unbiased in other areas. This problem is usually referred to as the tropical speed biases problem. Studies on this problem have mostly hypothesized physics-related causes, like convection, wrong height assignment for cirrus clouds, and gravitational waves, among others. In this work, however, we prove that the root cause of the tropical speed biases is algorithmic: a common misconception on the robustness of cross-correlation tracking, combined with an approximation made on the guess vector and the long temporal gap between consecutive images used from LEO satellites, is actually responsible not only for the positive biases observed in the tropics, but also for the negative biases at mid-latitudes. To demonstrate this, we will focus on the steps of the AMV tracking algorithm, including the use of the guess, the computation of the parallax correction and its effects depending on the wind direction and the relative position of the satellites at stake. As a consequence of this finding, EUMETSAT started to investigate new tracking algorithms independent of the wind guess to extract AMVs from LEO satellites.